# Active Retrieval

> Active retrieval augmented generation decides during generation when and what to retrieve; in FLARE, the model drafts the next sentence and, if it contains low-confidence tokens, uses that draft as a search query and regenerates the sentence with the retrieved evidence.

- Identifier: PTL-0136
- Category: Retrieval & Tool Use
- Canonical URL: https://protologue.com/t/active-retrieval/
- Also known as: FLARE, forward-looking active retrieval
- Introduced: 2023

## Description

Retrieving once before generation suits short answers but not long-form text whose information needs change as it is written. Triggering retrieval only on uncertainty limits extra calls to the points where the model is likely to be wrong.

## Broader terms

- [Retrieval-Augmented Generation](https://protologue.com/t/retrieval-augmented-generation/)

## Related terms

- [Log Probabilities](https://protologue.com/t/log-probabilities/)
- [Self-RAG](https://protologue.com/t/self-rag/)
- [Agentic RAG](https://protologue.com/t/agentic-rag/)

## Sources

- Jiang et al. (2023). Active Retrieval Augmented Generation. https://arxiv.org/abs/2305.06983

## Cite this entry

Protologue. (2026). Active Retrieval. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0136). https://protologue.com/t/active-retrieval/

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)